Running a kids' AI app at $1 per child a month
What it costs to run Peabody's AI for one child, and the budgets, caching and code-first choices that keep it there. Real limits, from the code.
Peabody (peabody.kids) limits the AI it spends on each child to $1 a month, and most children use far less. That limit is a number in the code, checked before every AI call, not a hope. Peabody is free through the end of 2026, so the budget is also what keeps a free product from running away with its own bill. This post lists the limits, says where the money goes, explains the five choices that keep spending low, and says plainly what Peabody has not measured yet.
The limit, in plain numbers
Every AI call Peabody makes is counted in dollars against the child it was for, and the lines below are the defaults in the code (lib/ai/policy.ts). Peabody is not a place that spends whatever a story happens to cost.
| Who | The most their AI use can cost |
|---|---|
| One child | $1 a month and $0.15 a day |
| One household | $0.60 a day |
| A visitor without an account | $0.25, ever |
| All visitors without accounts, together | $1 a day |
| All of Peabody | $5 a day and $40 a month; the hard stop is $45 |
Two more limits sit under the child's $1. A child can make at most 40 AI calls a day, not counting safety screens, which are never refused. And one feature is only a sub-line inside the child's dollar: the Ask Peabody box may use at most $0.25 of it. Whatever it does, it cannot use up the rest.
These are ceilings, not averages. A child who reads a story and does a few games will be nowhere near them.
Where the money goes: a story
Almost all of the AI spending is a story. A child taps ideas; an AI screen checks the ideas before anything is written; a writer model writes the story; screens check it after it is written; and a sample is reviewed for quality. Each step is a call that costs a little. Most of what a child does on Peabody does not call a model at all: games, practice, math lessons and what comes back for review are ordinary code.
Stories are also where the longest waits could be, which is why the opening comes first. The first pages are ready in seconds, and the rest of the story is shown right away. There is no daily limit on reading a story once it exists.
Why the limit holds: five choices
1. Code before a model.
1. Code before a model. The practice brain, which decides the next question and when something comes back, is ordinary code with published constants. So are the math lessons and the word spacing in a story. None of it spends a token. AI writes the stories; code decides the rest, and the code is free to run as often as a child plays.
2. Generate once, reuse. The stories in Peabody's library, and the quizzes for them, are made once and read by everyone. A safety verdict is cached only when the text is byte-for-byte the same, so a cached answer never stands in for a different story.
3. The opening first, then the rest. A story is written so that the first pages arrive quickly and the remainder follows. One tap is never a long wait, and a child who stops after the opening has not caused the whole story's work to be thrown away.
4. Reserve before every call, settle after. Before a call goes out, the gateway reserves budget against the child's lines. After it returns, the reservation is settled against what the call used. A call that cannot be counted is refused rather than allowed through uncounted. The one exception is deliberate: a safety screen for a story already being written is never refused, so a child can go over a line by a few cents of screening and never by leaving a story unchecked.
5. The cheapest model that meets the safety bar. The writer and the first-pass screens run on Claude Haiku 4.5. A second opinion on a screen, and the sampled quality review, use a stronger model, but only when they are needed.
What the limit protects
The point of a per-child limit is not only the bill. It decides what a child can lose.
The point of a per-child limit is not only the bill. It decides what a child can lose. If the chat or the Ask box could spend without limit, a child's bedtime story could be the thing that runs out. Because the ask box and the chat are sub-lines inside the same dollar, they can be used up without touching the part of the budget that writes a story. The budget protects the story first.
It also protects Peabody from its own mistakes. The all-of-Peabody lines, $5 a day and $40 a month, with a hard stop at $45, mean that a bug that calls a model in a loop ends at a known number instead of an invoice.
What we do not know yet
Peabody has not published a monthly cost for the average child, because it has not measured one. The limits above are what a child's AI use is allowed to cost, not what it does cost.
Peabody has not published a monthly cost for the average child, because it has not measured one. The limits above are what a child's AI use is allowed to cost, not what it does cost. Anyone who quotes you a monthly figure per child for an app like this without a measured average is guessing, and so would Peabody be.
What Peabody can say is narrower: the limits are enforced in the code, they are checked by a test suite, and every call is logged against a child, so the average will be measurable. When it is measured, this post will be updated with it. Nothing here covers revenue, profit or how many children use Peabody, and none of it is a promise about how Peabody will be priced after 2026.
You can read how the AI is used, and where it is never used, on the AI policy page, the rules for practice on the science page, and the plain facts on the facts page. To see what a story is like, start free or read a story from the library.
Questions parents ask
60 a day. That is the limit, not a measured average: Peabody has not measured a typical month yet and does not claim one.
How much does Peabody spend on AI for my child?
At most $1 a month and $0.15 a day, and a household at most $0.60 a day. That is the limit, not a measured average: Peabody has not measured a typical month yet and does not claim one.
Can the chat or the Ask box use up my child's stories?
No. The Ask box can use at most $0.25 of a child's $1. It sits inside the child's budget, so it can run out without taking the part that writes a story.
Does Peabody use AI to decide what my child learns next?
No. The next question, the level and when something comes back are decided by ordinary code with published rules. AI writes and screens stories; it never decides a child's level. There is more on the AI policy page, and a related post on AI safety by design for kids' apps.
Checked against how Peabody works on 7 November 2026. Written with help from AI. Peabody is a reading app, not a reading assessment: if you're worried about your child's reading, talk to their teacher. Questions, or something wrong? Write to hello@peabody.kids.